loki-mode 5.4.5 → 5.4.7

package/SKILL.md

@@ -3,7 +3,7 @@ name: loki-mode
 description: Multi-agent autonomous startup system. Triggers on "Loki Mode". Takes PRD to deployed product with zero human intervention. Requires --dangerously-skip-permissions flag.
 ---
 
-# Loki Mode v5.4.5
+# Loki Mode v5.4.7
 
 **You are an autonomous agent. You make decisions. You do not ask questions. You do not stop.**
 
@@ -241,4 +241,4 @@ Auto-detected or force with `LOKI_COMPLEXITY`:
 
 ---
 
-**v5.4.5 | Dashboard fix + Gemini stdin fix | ~245 lines core**
+**v5.4.7 | Gemini rate limit fallback to flash model | ~245 lines core**

package/VERSION

@@ -1 +1 @@
-5.4.5
+5.4.7

package/autonomy/run.sh

@@ -1378,10 +1378,9 @@ spawn_worktree_session() {
                 ;;
             gemini)
                 # Note: -p flag is DEPRECATED per gemini --help. Using positional prompt.
-                # Note: < /dev/null prevents Gemini from pausing on stdin
-                gemini --yolo --model "${PROVIDER_MODEL:-gemini-3-pro-preview}" \
-                    "Loki Mode: $task_prompt. Read .loki/CONTINUITY.md for context." \
-                    < /dev/null >> "$log_file" 2>&1
+                # Uses invoke_gemini helper for rate limit fallback to flash model
+                invoke_gemini "Loki Mode: $task_prompt. Read .loki/CONTINUITY.md for context." \
+                    >> "$log_file" 2>&1
                 ;;
             *)
                 log_error "Unknown provider: ${PROVIDER_NAME}"
@@ -1471,8 +1470,8 @@ Output ONLY the resolved file content with no conflict markers. No explanations.
                 resolution=$(codex exec --dangerously-bypass-approvals-and-sandbox "$conflict_prompt" 2>/dev/null)
                 ;;
             gemini)
-                # Note: -p flag is DEPRECATED per gemini --help. Using positional prompt.
-                resolution=$(gemini --yolo --model "${PROVIDER_MODEL:-gemini-3-pro-preview}" "$conflict_prompt" < /dev/null 2>/dev/null)
+                # Uses invoke_gemini_capture for rate limit fallback to flash model
+                resolution=$(invoke_gemini_capture "$conflict_prompt" 2>/dev/null)
                 ;;
             *)
                 log_error "Unknown provider: ${PROVIDER_NAME}"
@@ -1864,6 +1863,112 @@ EOF
     log_info "Loki directory initialized: .loki/"
 }
 
+#===============================================================================
+# Gemini Invocation with Rate Limit Fallback
+#===============================================================================
+
+# Invoke Gemini with automatic fallback to flash model on rate limit
+# Usage: invoke_gemini "prompt" [additional args...]
+# Returns: exit code from gemini CLI
+invoke_gemini() {
+    local prompt="$1"
+    shift
+
+    local model="${PROVIDER_MODEL:-gemini-3-pro-preview}"
+    local fallback="${PROVIDER_MODEL_FALLBACK:-gemini-3-flash-preview}"
+
+    # Create temp file for output to preserve streaming while checking for rate limit
+    local tmp_output
+    tmp_output=$(mktemp)
+
+    # Try primary model first
+    gemini --yolo --model "$model" "$prompt" "$@" < /dev/null 2>&1 | tee "$tmp_output"
+    local exit_code=${PIPESTATUS[0]}
+
+    # Check for rate limit in output
+    if [[ $exit_code -ne 0 ]] && grep -qiE "(rate.?limit|429|quota|resource.?exhausted)" "$tmp_output"; then
+        log_warn "Rate limit hit on $model, falling back to $fallback"
+        rm -f "$tmp_output"
+        gemini --yolo --model "$fallback" "$prompt" "$@" < /dev/null
+        exit_code=$?
+    else
+        rm -f "$tmp_output"
+    fi
+
+    return $exit_code
+}
+
+# Invoke Gemini and capture output (for variable assignment)
+# Usage: result=$(invoke_gemini_capture "prompt")
+# Falls back to flash model on rate limit
+invoke_gemini_capture() {
+    local prompt="$1"
+    shift
+
+    local model="${PROVIDER_MODEL:-gemini-3-pro-preview}"
+    local fallback="${PROVIDER_MODEL_FALLBACK:-gemini-3-flash-preview}"
+    local output
+
+    # Try primary model first
+    output=$(gemini --yolo --model "$model" "$prompt" "$@" < /dev/null 2>&1)
+    local exit_code=$?
+
+    # Check for rate limit in output
+    if [[ $exit_code -ne 0 ]] && echo "$output" | grep -qiE "(rate.?limit|429|quota|resource.?exhausted)"; then
+        log_warn "Rate limit hit on $model, falling back to $fallback" >&2
+        output=$(gemini --yolo --model "$fallback" "$prompt" "$@" < /dev/null 2>&1)
+    fi
+
+    echo "$output"
+}
+
+#===============================================================================
+# Copy Skill Files to Project Directory
+#===============================================================================
+
+copy_skill_files() {
+    # Copy skill files from the CLI package to the project's .loki/ directory.
+    # This makes the CLI self-contained - no need to install Claude Code skill separately.
+    # All providers (Claude, Gemini, Codex) use the same .loki/skills/ location.
+
+    local skills_src="$PROJECT_DIR/skills"
+    local skills_dst=".loki/skills"
+
+    if [ ! -d "$skills_src" ]; then
+        log_warn "Skills directory not found at $skills_src"
+        return 1
+    fi
+
+    # Create destination and copy skill files
+    mkdir -p "$skills_dst"
+
+    # Copy all skill markdown files
+    local copied=0
+    for skill_file in "$skills_src"/*.md; do
+        if [ -f "$skill_file" ]; then
+            cp "$skill_file" "$skills_dst/"
+            ((copied++))
+        fi
+    done
+
+    # Also copy SKILL.md to .loki/ and rewrite paths for workspace access
+    if [ -f "$PROJECT_DIR/SKILL.md" ]; then
+        # Rewrite skill paths from skills/ to .loki/skills/
+        sed -e 's|skills/00-index\.md|.loki/skills/00-index.md|g' \
+            -e 's|skills/model-selection\.md|.loki/skills/model-selection.md|g' \
+            -e 's|skills/quality-gates\.md|.loki/skills/quality-gates.md|g' \
+            -e 's|skills/testing\.md|.loki/skills/testing.md|g' \
+            -e 's|skills/troubleshooting\.md|.loki/skills/troubleshooting.md|g' \
+            -e 's|skills/production\.md|.loki/skills/production.md|g' \
+            -e 's|skills/parallel-workflows\.md|.loki/skills/parallel-workflows.md|g' \
+            -e 's|skills/providers\.md|.loki/skills/providers.md|g' \
+            -e 's|Read skills/|Read .loki/skills/|g' \
+            "$PROJECT_DIR/SKILL.md" > ".loki/SKILL.md"
+    fi
+
+    log_info "Copied $copied skill files to .loki/skills/"
+}
+
 #===============================================================================
 # Task Status Monitor
 #===============================================================================
@@ -3415,7 +3520,8 @@ build_prompt() {
     # Core autonomous instructions - NO questions, NO waiting, NEVER say done
     local autonomous_suffix="CRITICAL AUTONOMY RULES: 1) NEVER ask questions - just decide. 2) NEVER wait for confirmation - just act. 3) NEVER say 'done' or 'complete' - there's always more to improve. 4) NEVER stop voluntarily - if out of tasks, create new ones (add tests, optimize, refactor, add features). 5) Work continues PERPETUALLY. Even if PRD is implemented, find bugs, add tests, improve UX, optimize performance."
 
-    local sdlc_instruction="SDLC_PHASES_ENABLED: [$phases]. Execute ALL enabled phases. Log results to .loki/logs/. See SKILL.md for phase details."
+    # Skill files are always copied to .loki/skills/ for all providers
+    local sdlc_instruction="SDLC_PHASES_ENABLED: [$phases]. Execute ALL enabled phases. Log results to .loki/logs/. See .loki/SKILL.md for phase details. Skill modules at .loki/skills/."
 
     # Codebase Analysis Mode - when no PRD provided
     local analysis_instruction="CODEBASE_ANALYSIS_MODE: No PRD. FIRST: Analyze codebase - scan structure, read package.json/requirements.txt, examine README. THEN: Generate PRD at .loki/generated-prd.md. FINALLY: Execute SDLC phases."
@@ -3830,13 +3936,25 @@ if __name__ == "__main__":
 
             gemini)
                 # Gemini: Degraded mode - no stream-json, no agent tracking
-                # Using --model flag to specify model
-                # Note: < /dev/null prevents Gemini from pausing on stdin
-                echo "[loki] Gemini model: ${PROVIDER_MODEL:-gemini-3-pro-preview}, tier: $tier_param" >> "$log_file"
-                echo "[loki] Gemini model: ${PROVIDER_MODEL:-gemini-3-pro-preview}, tier: $tier_param" >> "$agent_log"
-                gemini --yolo --model "${PROVIDER_MODEL:-gemini-3-pro-preview}" \
-                    "$prompt" < /dev/null 2>&1 | tee -a "$log_file" "$agent_log"
+                # Uses invoke_gemini helper for rate limit fallback to flash model
+                local model="${PROVIDER_MODEL:-gemini-3-pro-preview}"
+                local fallback="${PROVIDER_MODEL_FALLBACK:-gemini-3-flash-preview}"
+                echo "[loki] Gemini model: $model (fallback: $fallback), tier: $tier_param" >> "$log_file"
+                echo "[loki] Gemini model: $model (fallback: $fallback), tier: $tier_param" >> "$agent_log"
+
+                # Try primary model, fallback on rate limit
+                local tmp_output
+                tmp_output=$(mktemp)
+                gemini --yolo --model "$model" "$prompt" < /dev/null 2>&1 | tee "$tmp_output" | tee -a "$log_file" "$agent_log"
                 local exit_code=${PIPESTATUS[0]}
+
+                if [[ $exit_code -ne 0 ]] && grep -qiE "(rate.?limit|429|quota|resource.?exhausted)" "$tmp_output"; then
+                    log_warn "Rate limit hit on $model, falling back to $fallback"
+                    echo "[loki] Fallback to $fallback due to rate limit" >> "$log_file"
+                    gemini --yolo --model "$fallback" "$prompt" < /dev/null 2>&1 | tee -a "$log_file" "$agent_log"
+                    exit_code=${PIPESTATUS[0]}
+                fi
+                rm -f "$tmp_output"
                 ;;
 
             *)
@@ -4276,6 +4394,10 @@ main() {
     # Initialize .loki directory
     init_loki_dir
 
+    # Copy skill files to .loki/skills/ - makes CLI self-contained
+    # No need to install Claude Code skill separately
+    copy_skill_files
+
     # Import GitHub issues if enabled (v4.1.0)
     if [ "$GITHUB_IMPORT" = "true" ]; then
         import_github_issues

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "loki-mode",
-  "version": "5.4.5",
+  "version": "5.4.7",
   "description": "Multi-agent autonomous startup system for Claude Code, Codex CLI, and Gemini CLI",
   "keywords": [
     "claude",

package/providers/gemini.sh

@@ -49,11 +49,13 @@ PROVIDER_MAX_PARALLEL=1
 
 # Model Configuration
 # Gemini CLI supports --model flag to specify model
-# Using gemini-3-pro-preview (latest as of Jan 2026)
+# Primary: gemini-3-pro-preview (latest as of Jan 2026)
+# Fallback: gemini-3-flash-preview (for rate limit scenarios)
 PROVIDER_MODEL="gemini-3-pro-preview"
+PROVIDER_MODEL_FALLBACK="gemini-3-flash-preview"
 PROVIDER_MODEL_PLANNING="gemini-3-pro-preview"
 PROVIDER_MODEL_DEVELOPMENT="gemini-3-pro-preview"
-PROVIDER_MODEL_FAST="gemini-3-pro-preview"
+PROVIDER_MODEL_FAST="gemini-3-flash-preview"
 
 # Thinking levels (Gemini-specific: maps to reasoning depth)
 PROVIDER_THINKING_PLANNING="high"
@@ -96,14 +98,28 @@ provider_version() {
     gemini --version 2>/dev/null | head -1
 }
 
-# Invocation function
+# Invocation function with rate limit fallback
 # Uses --model flag to specify model, --yolo for autonomous mode
-# Using positional prompt (not deprecated -p flag)
+# Falls back to flash model if pro hits rate limit
 # Note: < /dev/null prevents Gemini from pausing on stdin
 provider_invoke() {
     local prompt="$1"
     shift
-    gemini --yolo --model "$PROVIDER_MODEL" "$prompt" "$@" < /dev/null
+    local output
+    local exit_code
+
+    # Try primary model first
+    output=$(gemini --yolo --model "$PROVIDER_MODEL" "$prompt" "$@" < /dev/null 2>&1)
+    exit_code=$?
+
+    # Check for rate limit (429) or quota exceeded
+    if [[ $exit_code -ne 0 ]] && echo "$output" | grep -qiE "(rate.?limit|429|quota|resource.?exhausted)"; then
+        echo "[loki] Rate limit hit on $PROVIDER_MODEL, falling back to $PROVIDER_MODEL_FALLBACK" >&2
+        gemini --yolo --model "$PROVIDER_MODEL_FALLBACK" "$prompt" "$@" < /dev/null
+    else
+        echo "$output"
+        return $exit_code
+    fi
 }
 
 # Model tier to thinking level parameter
@@ -117,14 +133,34 @@ provider_get_tier_param() {
     esac
 }
 
-# Tier-aware invocation
+# Tier-aware invocation with rate limit fallback
 # Uses --model flag to specify model
-# Using positional prompt (not deprecated -p flag)
+# Falls back to flash model if pro hits rate limit
 # Note: < /dev/null prevents Gemini from pausing on stdin
 provider_invoke_with_tier() {
     local tier="$1"
     local prompt="$2"
     shift 2
-    echo "[loki] Using tier: $tier, model: $PROVIDER_MODEL" >&2
-    gemini --yolo --model "$PROVIDER_MODEL" "$prompt" "$@" < /dev/null
+
+    # Select model based on tier
+    local model="$PROVIDER_MODEL"
+    [[ "$tier" == "fast" ]] && model="$PROVIDER_MODEL_FAST"
+
+    echo "[loki] Using tier: $tier, model: $model" >&2
+
+    local output
+    local exit_code
+
+    # Try selected model first
+    output=$(gemini --yolo --model "$model" "$prompt" "$@" < /dev/null 2>&1)
+    exit_code=$?
+
+    # Check for rate limit (429) or quota exceeded - fallback to flash
+    if [[ $exit_code -ne 0 ]] && echo "$output" | grep -qiE "(rate.?limit|429|quota|resource.?exhausted)"; then
+        echo "[loki] Rate limit hit on $model, falling back to $PROVIDER_MODEL_FALLBACK" >&2
+        gemini --yolo --model "$PROVIDER_MODEL_FALLBACK" "$prompt" "$@" < /dev/null
+    else
+        echo "$output"
+        return $exit_code
+    fi
 }